A system and method for automatically reviewing video AI recognition results

Abstract

The present invention discloses a system and method for automatically reviewing video AI recognition results, and relates to the field of video processing. Specifically disclosed is a system for automatically reviewing video AI recognition results, including: a video AI analysis module, which performs AI analysis on the video screen to obtain a rectangular frame of the target; a front detection module, which detects the target position of the video screen; a picture adjustment module, which calculates and adjusts the camera so that the target is in the middle of the video screen; a review and determination module, which selects the final recognition result through multiple captured video screens. The present invention can reduce the misjudgment caused by video quality factors, thereby effectively reducing the workload of manual review, and improving the overall performance and user experience of video AI analysis technology in industrial production safety applications.

Description

Technical Field

[0001] The present invention relates to the field of video processing, and in particular to a system and method for automatically reviewing video AI recognition results. Background Art

[0002] Video AI analysis technology mainly relies on advanced industrial machine vision algorithm framework to achieve in-depth analysis and insight into video data. Its application scope covers a wide range of industrial production safety fields. This technology can efficiently identify and monitor unsafe operating behaviors of personnel, the operating status of industrial equipment, and potential safety hazards in the industrial environment. Through an immediate warning mechanism, it can significantly improve the safety and efficiency of production operations. However, although video AI analysis technology has made significant progress, it still faces challenges in the pursuit of absolute accuracy and cannot achieve accurate recognition. Therefore, in the actual deployment and application process, in order to ensure the reliability and effectiveness of the analysis results, manual review has become an indispensable link. At present, in the case of algorithm misjudgment, a significant problem comes from the limitation of video quality. Specifically, the target object in the camera's field of view is partially blocked due to being too far away or at the edge of the picture, resulting in unclear image information, which in turn causes algorithm misjudgment. Such misjudgment not only affects the accuracy of the analysis results, but also invisibly increases the burden of manual review and reduces overall work efficiency. In view of the above situation, the current field urgently needs a system and method for automatically reviewing video AI recognition results to achieve innovative solutions, aiming to reduce misjudgments caused by video quality factors, thereby effectively reducing the workload of manual review and improving the overall performance and user experience of video AI analysis technology in industrial production safety applications. Summary of the invention

[0003] The technical problem to be solved by the present invention is how to implement innovative solutions, reduce misjudgments caused by video quality factors, thereby effectively reducing the workload of manual review, and improving the overall performance and user experience of video AI analysis technology in industrial production safety applications. The purpose is to provide a system and method for automatically reviewing video AI recognition results, which solves the above-mentioned technical problems.

[0004] The present invention is achieved through the following technical solutions:

[0005] A system for automatically reviewing video AI recognition results, comprising:

[0006] The video AI analysis module performs AI analysis on the video image to obtain the target rectangle;

[0007] The front detection module detects the target position of the video screen;

[0008] The picture adjustment module calculates and adjusts the camera to place the target in the middle of the video picture;

[0009] The review and judgment module selects the final recognition result through multiple captured video images.

[0010] The above-mentioned pre-detection module includes a real-time verification module, a morphological anomaly analysis module, a position anomaly analysis module and a distance anomaly analysis module;

[0011] The real-time verification module is used to determine whether the delay time of the video screen exceeds the maximum delay threshold; the maximum delay threshold is lower than the time setting for the target to move out of the video screen;

[0012] When the above-mentioned maximum delay threshold is exceeded, the video delay time is determined to be abnormal;

[0013] When the maximum delay threshold is not exceeded, the video delay time is determined to be normal, and the morphological abnormality analysis module determines whether the deviation between the aspect ratio of the target and the actual aspect ratio of the target reaches an aspect ratio deviation threshold; the aspect ratio deviation threshold is set according to the ratio of the size of the target and the actual target size;

[0014] When the aspect ratio deviation threshold is reached, the target is determined to be occluded;

[0015] When it is lower than the aspect ratio deviation threshold, the position anomaly analysis module determines whether the distance from the target to the center of the video screen reaches the center distance threshold; when it reaches the center distance threshold, it is determined that the target is at the edge of the screen; the center distance threshold is set according to the size of the video screen;

[0016] When the distance is lower than the center distance threshold, the distance anomaly analysis module determines whether the ratio of the size of the target to the video screen reaches the size ratio threshold; the size ratio threshold is set according to the distance from the target to the center of the video screen;

[0017] When the size ratio exceeds the above threshold, it is determined that the distance to the target is too far; when the size ratio is below the above threshold, it is determined that the distance to the target is normal.

[0018] The above-mentioned picture adjustment module includes:

[0019] A real-time adjustment module, when determining that the video delay time is abnormal, performs real-time calibration on the video image so that the delay time of the video image does not exceed the above-mentioned maximum delay threshold;

[0020] The occlusion adjustment module, when determining that the target is occluded, compares the shape of the target with the actual shape of the target to obtain the occlusion area, moves the camera away from the occlusion area and adjusts the rotation direction so that the deviation between the aspect ratio of the target and the actual aspect ratio of the target is lower than the aspect ratio deviation threshold;

[0021] The edge adjustment module, when determining that the target is at the edge of the screen, rotates the camera toward the target so that the distance from the target to the center of the video screen does not exceed the center distance threshold;

[0022] The distance adjustment module, when determining that the distance of the target is too far, moves the camera toward the target so that the ratio of the size of the target to the video screen meets the above-mentioned size ratio threshold, or adjusts the focal length of the camera to improve the resolution.

[0023] The steps of adjusting the focal length of the camera to improve the resolution include:

[0024] Calculate the zoom ratio of the target size to the ideal size;

[0025] According to the optical characteristics of the camera, the relationship between the focal length and the zoom ratio is obtained;

[0026] Adjust the focal length according to the zoom ratio.

[0027] The above-mentioned step of selecting the final recognition result through multiple captured video images includes:

[0028] After the above-mentioned picture adjustment module adjusts the video picture, the majority is taken according to the recognition results of the single video pictures to obtain the final recognition result.

[0029] The above-mentioned system for automatically reviewing video AI recognition results also includes:

[0030] Multiple cameras to collect video images of the target;

[0031] Video acquisition training module: perform feature analysis based on the final recognition results of multiple cameras to obtain the video screen with the most recognition features; collect multiple groups of video acquisition training data; each group of the above video acquisition training data includes video screens acquired by multiple cameras, and the selected video screen with the most recognition features; obtain a video acquisition training model through machine learning training of the multiple groups of the above video acquisition training data; output the selected video screen according to the video screen to be tested; and provide the selected video screen to the above video AI analysis module.

[0032] The above-mentioned system for automatically reviewing video AI recognition results also includes:

[0033] The video adjustment training module collects multiple sets of video adjustment training data, each set of the above video adjustment training data includes video images and camera adjustment parameters; the multiple sets of the above video adjustment training data obtain video adjustment training models through machine learning; the camera adjustment parameters of the test images are obtained through the above video adjustment training model; the above camera adjustment parameters include any one or more of the camera's focal length, horizontal rotation angle, vertical rotation angle and translation vector.

[0034] A method for automatically reviewing video AI recognition results, comprising:

[0035] Perform AI analysis on the video to get the rectangular frame of the target;

[0036] Detect the target position of the video screen;

[0037] Calculate and adjust the camera so that the target is in the center of the video frame;

[0038] The final recognition result is selected through multiple captured video images.

[0039] An electronic device comprises a memory, a processor and a computer program running on the processor, wherein the processor implements the above-mentioned system for automatically reviewing video AI recognition results when executing the computer program.

[0040] A computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, wherein when the computer program is executed by a processor, the system for automatically reviewing video AI recognition results is implemented.

[0041] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0042] The present invention proposes a method for automatically reviewing the AI ​​recognition results of industrial videos. The video screen is analyzed by AI technology to obtain a rectangular frame of the target. After the target is detected, the camera is calculated and adjusted to place the target in the middle of the video screen, thereby accurately tracking the target, and finally selecting the rectangular frame of the target by using the video screen captured multiple times, thereby reducing the misjudgment caused by video quality factors, thereby effectively reducing the workload of manual review, and improving the overall performance and user experience of video AI analysis technology in industrial production safety applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, other relevant drawings can be obtained based on these drawings without creative work. In the drawings:

[0044] Figure 1 This is a flowchart of a system for automatically reviewing video AI recognition results according to an embodiment of the present application. DETAILED DESCRIPTION

[0045] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0046] Example

[0047] like Figure 1 As shown, the embodiment of the present application provides a system for automatically reviewing video AI recognition results, including:

[0048] The video AI analysis module performs AI analysis on the video image to obtain the target rectangle;

[0049] The front detection module detects the target position of the video screen;

[0050] The picture adjustment module calculates and adjusts the camera to place the target in the middle of the video picture;

[0051] The review and judgment module selects the final recognition result through multiple captured video images.

[0052] The above-mentioned pre-detection module includes a real-time verification module, a morphological anomaly analysis module, a position anomaly analysis module and a distance anomaly analysis module;

[0053] The real-time verification module is used to determine whether the delay time of the video screen exceeds the maximum delay threshold; the maximum delay threshold is lower than the time setting for the target to move out of the video screen;

[0054] When the above-mentioned maximum delay threshold is exceeded, the video delay time is determined to be abnormal;

[0055] When the maximum delay threshold is not exceeded, the video delay time is determined to be normal, and the morphological abnormality analysis module determines whether the deviation between the aspect ratio of the target and the actual aspect ratio of the target reaches an aspect ratio deviation threshold; the aspect ratio deviation threshold is set according to the ratio of the size of the target and the actual target size;

[0056] When the aspect ratio deviation threshold is reached, the target is determined to be occluded;

[0057] When it is lower than the aspect ratio deviation threshold, the position anomaly analysis module determines whether the distance from the target to the center of the video screen reaches the center distance threshold; when it reaches the center distance threshold, it is determined that the target is at the edge of the screen; the center distance threshold is set according to the size of the video screen;

[0058] When the distance is lower than the center distance threshold, the distance anomaly analysis module determines whether the ratio of the size of the target to the video screen reaches the size ratio threshold; the size ratio threshold is set according to the distance from the target to the center of the video screen;

[0059] When the size ratio exceeds the above threshold, it is determined that the distance to the target is too far; when the size ratio is below the above threshold, it is determined that the distance to the target is normal.

[0060] The above AI technology is an existing technology that can identify dynamic or static features of images in video images, such as the target's face, clothing features (such as clothes, hats and glasses, etc.) and body movements, any one feature or a combination of multiple features. When the video AI analysis module has a recognition result, the rectangular area information of the target is pushed to the front detection module, and the location of the recognition result is intelligently detected to determine the need for automatic review.

[0061] If the delay of the video screen is too large, the original target may no longer exist during the review by the review judgment module. In order to ensure the timeliness of the automatic review, the delay time of the video screen is set lower than the time threshold of the target moving in the current video screen. If the threshold is exceeded, no review will be performed.

[0062] Identify whether the deviation between the aspect ratio of the target and the common aspect ratio in the historical target data exceeds the threshold. If it exceeds the threshold, the target is judged to be suspected of being occluded. Specifically, when using historical data to calculate the common aspect ratio range, the target aspect ratio can be expressed as AR=Hobj / Wobj. If |AR−μAR|>σAR×τAR (μAR is the average aspect ratio, σAR is the standard deviation, and τAR is the deviation threshold), the target is judged to be suspected of being occluded.

[0063] Identify the distance from the center point of the target to the center point of the entire screen. When the distance to the center of the screen exceeds the threshold, the target is determined to be at the edge of the screen. The threshold can be set according to the size of the video screen. When the video screen is larger, the actual proportion of the target is smaller, and the threshold is set smaller. For example, when the target is the center of gravity of the video screen, the ratio of the distance between the target and the center of the screen and the width of the video screen should not exceed the ratio of the actual area of ​​the target to the actual area of ​​the video screen, so the threshold is set to the product of the actual proportion of the target and the video and the width of the screen. Assume that the coordinates of the center point of the target are (xobj, yobj) and the center point of the screen is (Wframe / 2, Hframe / 2), then the distance D is:

[0064] ;

[0065] If P≤τpos (τpos is a preset threshold), it is determined that the target is at the edge of the screen.

[0066] Furthermore, when the highest requirement of the center distance is met, the size of the target meets the minimum requirement of the screen size. The ratio of the size of the target to the size of the screen is set according to the distance between the current target and the center: when the target is closer to the center, the closer the target is to the camera, the larger the size, and vice versa. The threshold can be specifically set according to the size of the reference object that needs to be collected in the video screen.

[0067] When all the above detection contents are met, it will be determined to meet the conditions for automatic review and enter the picture adjustment module for processing.

[0068] The above-mentioned picture adjustment module includes:

[0069] A real-time adjustment module, when determining that the video delay time is abnormal, performs real-time calibration on the video image so that the delay time of the video image does not exceed the above-mentioned maximum delay threshold;

[0070] The occlusion adjustment module, when determining that the target is occluded, compares the shape of the target with the actual shape of the target to obtain the occlusion area, moves the camera away from the occlusion area and adjusts the rotation direction so that the deviation between the aspect ratio of the target and the actual aspect ratio of the target is lower than the aspect ratio deviation threshold;

[0071] The edge adjustment module, when determining that the target is at the edge of the screen, rotates the camera toward the target so that the distance from the target to the center of the video screen does not exceed the center distance threshold;

[0072] The distance adjustment module, when determining that the distance of the target is too far, moves the camera toward the target so that the ratio of the size of the target to the video screen meets the above-mentioned size ratio threshold, or adjusts the focal length of the camera to improve the resolution.

[0073] The steps of adjusting the focal length of the camera to improve the resolution include:

[0074] Calculate the zoom ratio of the target size to the ideal size;

[0075] According to the optical characteristics of the camera, the relationship between the focal length and the zoom ratio is obtained;

[0076] Adjust the focal length according to the zoom ratio.

[0077] In order to ensure that the target to be reviewed is clearly located in the center of the picture, adjust the camera to perform operations such as zooming, panning, and possible rotation.

[0078] First, define a camera parameter model, including focal length f, horizontal rotation angle θ, vertical rotation angle ϕ, and translation vector p=(px,py) (for cameras that support translation). These parameters together determine the viewing angle and position of the camera's captured image.

[0079] Assume that the position of the target in the picture is (xobj, yobj) and the center of the picture is (Wframe / 2, Hframe / 2). Then the deviation vector between the target position and the center of the picture is for:

[0080]

[0081] For cameras that support zoom, the following algorithm is designed to adjust the focal length f so that the target occupies a more appropriate proportion in the picture:

[0082] Calculate the ratio of the target size to the ideal size: Assume the current size of the target is (Wobj, Hobj) and the ideal size is (Wideal, Hideal) (based on the screen size and the preset display ratio), then the zoom ratio k is:

[0083]

[0084] Adjusting focal length: According to the optical characteristics of the camera, the relationship between focal length f and zoom ratio k may be linear, nonlinear, or based on a specific function, and is adjusted according to actual conditions. For example, a case in implementation is a linear relationship:

[0085]

[0086] Where fold is the focal length before adjustment, and fnew is the focal length after adjustment.

[0087] According to the deviation vector Calculate the translation amount. Since the camera's translation capability may be limited, and the distance of each movement may be fixed or adjustable, a method of multiple movements is designed. Assuming that the maximum single movable distance of the camera is (Δxmax, Δymax), the single movement amount (Δx, Δy) can be a proportion or fixed value of d, but not exceeding the maximum movable distance:

[0088] ;

[0089] in and They are The horizontal and vertical components of the , sign() function is used to determine the direction of movement.

[0090] Iterative movement: If one movement is not enough to move the target to the center of the picture, the above process is repeated until the target position meets the preset conditions (such as the distance from the center of the picture is less than a certain threshold).

[0091] When the target position has a large deviation or is at the edge, you may consider rotating it diagonally. Optionally, the rotation angles θ and ϕ are determined based on the target position, the camera position, and the camera's rotation axis. They are usually calculated based on the angle between the target position and the center of the image, and the camera's rotation angle is adjusted accordingly.

[0092] Through the above method, the picture adjustment module can realize the control of the camera viewing angle and provide high-quality video pictures for subsequent review and judgment.

[0093] The above-mentioned step of selecting the final recognition result through multiple captured video images includes:

[0094] After the above-mentioned picture adjustment module adjusts the video picture, the majority is taken according to the recognition results of the single video pictures to obtain the final recognition result.

[0095] When the picture adjustment module completes the picture adjustment, the review and judgment module will start capturing pictures and send them to the video AI analysis module for recognition. For the recognition results, an odd-number voting strategy is adopted, that is, the final result is determined by a majority election mechanism. For example, in the case of capturing 2 pictures, combined with the first recognition result, there are a total of 3 results. If 2 are alarms, then it is judged as an alarm, and vice versa. Assume that the number of captures is N, and the recognition results are {R0, R1, R2,..., RN} (where R0 is the initial recognition result), then the final judgment result F is:

[0096] F = mode({R0,R1,R2,...,RN})

[0097] The mode() function represents taking the majority. If the majority is an alarm, it is determined to be an alarm.

[0098] After completing the automatic review of the video AI analysis and recognition results, the camera will be notified to return to its original position and return to normal monitoring status.

[0099] The above-mentioned system for automatically reviewing video AI recognition results also includes:

[0100] Multiple cameras to collect video images of the target;

[0101] Video acquisition training module: perform feature analysis based on the final recognition results of multiple cameras to obtain the video screen with the most recognition features; collect multiple groups of video acquisition training data; each group of the above video acquisition training data includes video screens acquired by multiple cameras, and the selected video screen with the most recognition features; obtain a video acquisition training model through machine learning training of the multiple groups of the above video acquisition training data; output the selected video screen according to the video screen to be tested; and provide the selected video screen to the above video AI analysis module.

[0102] Factors such as rapid movement, occlusion, and poor transmission signals can cause inconsistencies in video capture and recognition. Therefore, machine learning can screen out the optimal solution, which is then fed back to video capture to obtain the optimal images captured by multiple cameras.

[0103] The above-mentioned system for automatically reviewing video AI recognition results also includes:

[0104] The video adjustment training module collects multiple sets of video adjustment training data, each set of the above video adjustment training data includes video images and camera adjustment parameters; the multiple sets of the above video adjustment training data obtain video adjustment training models through machine learning; the camera adjustment parameters of the test images are obtained through the above video adjustment training model; the above camera adjustment parameters include any one or more of the camera's focal length, horizontal rotation angle, vertical rotation angle and translation vector.

[0105] Through the video adjustment training model, a comprehensive camera adjustment strategy that supports zoom, translation and rotation is obtained. By weighing and optimizing multiple adjustment parameters, the camera parameters are adjusted in the best way, so that the target can be moved quickly and accurately to the center of the picture, and the picture clarity meets the requirements.

[0106] In summary, the present application provides a system and method for automatically reviewing video AI recognition results:

[0107] The pre-detection module is used to detect the recognition results and determine whether automatic review is needed; the picture adjustment module calculates and adjusts the camera to make the video picture in a better state; the review and judgment module makes judgments through multiple captures and selects the final result. It can automatically review the situation where the algorithm misjudges the target object within the camera's field of view because it is too far away or at the edge of the picture and is partially blocked, thereby effectively reducing the workload of manual review and improving the overall performance and user experience of video AI analysis technology in industrial production safety applications.

[0108] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A system for automatically reviewing video AI recognition results, characterized in that: include: The video AI analysis module performs AI analysis on the video image to obtain the target rectangle; The front detection module detects the target position of the video screen; The pre-detection module includes a real-time verification module, a morphological anomaly analysis module, a position anomaly analysis module and a distance anomaly analysis module; The real-time verification module is used to determine whether the delay time of the video screen exceeds the maximum delay threshold; the maximum delay threshold is lower than the time setting for the target to move out of the video screen; When the maximum delay threshold is exceeded, determining that the video delay time is abnormal; When the maximum delay threshold is not exceeded, the video delay time is determined to be normal, and the morphological abnormality analysis module determines whether the deviation between the aspect ratio of the target and the actual aspect ratio of the target reaches an aspect ratio deviation threshold; the aspect ratio deviation threshold is set according to the ratio of the size of the target to the actual target size; When the aspect ratio deviation threshold is reached, determining that the target is blocked; When it is lower than the aspect ratio deviation threshold, the position anomaly analysis module determines whether the distance from the target to the center of the video screen reaches the center distance threshold; when it reaches the center distance threshold, it is determined that the target is at the edge of the screen; the center distance threshold is set according to the size of the video screen; When the distance is lower than the center distance threshold, the distance anomaly analysis module determines whether the ratio of the size of the target to the video screen reaches a size ratio threshold; the size ratio threshold is set according to the distance from the target to the center of the video screen; When the size ratio threshold is exceeded, it is determined that the target is too far away; When the size ratio is below the threshold, the target distance is determined to be normal; The picture adjustment module calculates and adjusts the camera to place the target in the middle of the video picture; The review and judgment module selects the final recognition result through multiple captured video images.

2. A system for automatically reviewing video AI recognition results according to claim 1, characterized in that: The picture adjustment module comprises: A real-time adjustment module, when determining that the video delay time is abnormal, performs real-time calibration on the video picture so that the delay time of the video picture does not exceed the maximum delay threshold; The occlusion adjustment module compares the shape of the target with the actual shape of the target to obtain the occlusion area when determining that the target is occluded, moves the camera to avoid the occlusion area and adjusts the rotation direction so that the deviation between the aspect ratio of the target and the actual aspect ratio of the target is lower than the aspect ratio deviation threshold; The edge adjustment module, when determining that the target is at the edge of the screen, rotates the camera toward the target so that the distance from the target to the center of the video screen does not exceed the center distance threshold; The distance adjustment module, when determining that the distance of the target is too far, moves the camera toward the target so that the ratio of the size of the target to the video screen meets the size ratio threshold, or adjusts the focal length of the camera to improve the resolution.

3. A system for automatically reviewing video AI recognition results according to claim 2, characterized in that: The step of adjusting the focal length of the camera to improve the resolution includes: Calculate the zoom ratio of the target size to the ideal size; According to the optical characteristics of the camera, the relationship between the focal length and the zoom ratio is obtained; Adjust the focal length according to the zoom ratio.

4. The system for automatically reviewing video AI recognition results according to claim 1, characterized in that: The step of selecting the final recognition result through multiple captured video images includes: After the picture adjustment module adjusts the video picture, the majority is taken according to the recognition results of the single video pictures to obtain the final recognition result.

5. The system for automatically reviewing video AI recognition results according to claim 1, characterized in that: Also includes: Multiple cameras to collect video images of the target; Video acquisition training module: Perform feature analysis based on the final recognition results of multiple cameras to obtain the video images with the most recognition features; Collect multiple sets of video collection training data; Each set of the video acquisition training data includes video images acquired by multiple cameras, and the video images with the most identification features selected; A plurality of groups of the video acquisition training data are trained through machine learning to obtain a video acquisition training model; a selected video picture is output according to the video picture to be tested; and the selected video picture is provided to the video AI analysis module.

6. The system for automatically reviewing video AI recognition results according to claim 1, characterized in that: Also includes: A video adjustment training module collects multiple sets of video adjustment training data, each set of the video adjustment training data includes a video screen and camera adjustment parameters; Multiple groups of video adjustment training data are used to obtain video adjustment training models through machine learning; camera adjustment parameters of the test picture are obtained through the video adjustment training model; the camera adjustment parameters include any one or more of the camera's focal length, horizontal rotation angle, vertical rotation angle and translation vector.

7. A method for automatically reviewing video AI recognition results, characterized in that: include: Perform AI analysis on the video to get the rectangular frame of the target; Detect the target position of the video screen, including: Determine whether the delay time of the video screen exceeds a maximum delay threshold; the maximum delay threshold is lower than the time setting for the target to move out of the video screen; When the maximum delay threshold is exceeded, determining that the video delay time is abnormal; When the maximum delay threshold is not exceeded, the video delay time is determined to be normal, and whether the deviation between the aspect ratio of the target and the actual aspect ratio of the target reaches an aspect ratio deviation threshold is determined; the aspect ratio deviation threshold is set according to the ratio of the size of the target to the actual target size; When the aspect ratio deviation threshold is reached, determining that the target is blocked; When it is lower than the aspect ratio deviation threshold, it is determined whether the distance from the target to the center of the video screen reaches the center distance threshold; when it reaches the center distance threshold, it is determined that the target is at the edge of the screen; the center distance threshold is set according to the size of the video screen; When the distance is lower than the center distance threshold, it is determined whether the ratio of the size of the target to the video screen reaches a size ratio threshold; the size ratio threshold is set according to the distance from the target to the center of the video screen; When the size ratio exceeds the threshold, it is determined that the distance to the target is too far; when the size ratio is below the threshold, it is determined that the distance to the target is normal; Calculate and adjust the camera so that the target is in the center of the video frame; The final recognition result is selected through multiple captured video images.

8. An electronic device comprising a memory, a processor, and a computer program running on the processor, characterized in that: When the processor executes the computer program, it implements a method for automatically reviewing video AI recognition results as described in claim 7.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by the processor, it implements the method for automatically reviewing the video AI recognition results as described in claim 7.

Images